> If you're using AI, a language with a large training set is going to win

Not necessarily? What if the training set contains an overwhelming amount if bad code written by neophytes? I imagine Python quality by the LLM suffers from this, for example.

What if the language has extremely confusing syntax constructs (like early php) or bad or no conventions (suppose the standard library has somecollection.put(key, value) sometimes and othercollection.put(value, key) other times), and individual code authors just pick what they want adhoc

Large training set ain't gonna save you.